Hawk Lee
e21a5c9eba
The eff537d commit changed all files from 755 to 644.
Restores original executable permissions.
89 lines
3.0 KiB
Python
Executable File
89 lines
3.0 KiB
Python
Executable File
import sys
|
|
import os
|
|
import torch
|
|
import torchaudio
|
|
|
|
# Explicitly add remote paths
|
|
base_dir = "/app/ComfyUI/custom_nodes/ComfyUI_AIIA"
|
|
sys.path.insert(0, os.path.join(base_dir, "libs/CosyVoice"))
|
|
sys.path.insert(0, os.path.join(base_dir, "libs/CosyVoice/third_party/Matcha-TTS"))
|
|
|
|
print(f"Debug: sys.path[0-2]: {sys.path[:3]}")
|
|
|
|
try:
|
|
from cosyvoice.cli.cosyvoice import CosyVoice
|
|
print("Success: Imported CosyVoice")
|
|
except ImportError as e:
|
|
print(f"Error importing CosyVoice: {e}")
|
|
# Listing dir to see what's wrong
|
|
cv_path = os.path.join(base_dir, "libs/CosyVoice")
|
|
print(f"Listing {cv_path}: {os.listdir(cv_path) if os.path.exists(cv_path) else 'Not Found'}")
|
|
sys.exit(1)
|
|
|
|
def test_instruct():
|
|
# Try multiple possible model locations
|
|
possible_paths = [
|
|
"/app/ComfyUI/models/cosyvoice/CosyVoice-300M-Instruct",
|
|
"/app/ComfyUI/models/cosyvoice/FunAudioLLM/CosyVoice-300M-Instruct",
|
|
"pretrained_models/CosyVoice-300M-Instruct",
|
|
os.path.join(base_dir, "libs/CosyVoice/pretrained_models/CosyVoice-300M-Instruct")
|
|
]
|
|
|
|
model_dir = None
|
|
for p in possible_paths:
|
|
if os.path.exists(p):
|
|
model_dir = p
|
|
break
|
|
|
|
if not model_dir:
|
|
print("Error: Could not find CosyVoice-300M-Instruct model.")
|
|
print(f"Checked: {possible_paths}")
|
|
return
|
|
|
|
print(f"Loading model from {model_dir}")
|
|
|
|
cosyvoice = CosyVoice(model_dir)
|
|
print("Model loaded.")
|
|
|
|
# List of test cases for Dialect Prompt Tuning
|
|
test_cases = [
|
|
{
|
|
"name": "Dialect_01_CN_Term_Cantonese",
|
|
"text": "雷好啊，今晚去边度食饭？",
|
|
"spk": "中文女",
|
|
"instruct": "A female speaker with a 广东话 accent.<|endofprompt|>"
|
|
},
|
|
{
|
|
"name": "Dialect_02_CN_Instruction_Cantonese",
|
|
"text": "雷好啊，今晚去边度食饭？",
|
|
"spk": "中文女",
|
|
"instruct": "A female speaker. 用广东话。<|endofprompt|>"
|
|
},
|
|
{
|
|
"name": "Dialect_03_Sichuan_Native_Term",
|
|
"text": "我们四川人说话就是这个味儿。",
|
|
"spk": "中文男",
|
|
"instruct": "A male speaker with a 四川话 accent.<|endofprompt|>"
|
|
}
|
|
]
|
|
|
|
for case in test_cases:
|
|
print(f"Generating Case: {case['name']}")
|
|
print(f" Instruct: {case['instruct']}")
|
|
try:
|
|
output = cosyvoice.inference_instruct(case['text'], case['spk'], case['instruct'])
|
|
for i, j in enumerate(output):
|
|
# Ensure filename is safe
|
|
safe_name = case['name'].replace(" ", "_").replace("/", "-")
|
|
filename = f"sample_{safe_name}.wav"
|
|
abs_save_path = os.path.join(base_dir, filename)
|
|
torchaudio.save(abs_save_path, j['tts_speech'], cosyvoice.sample_rate)
|
|
print(f" Saved to: {abs_save_path}")
|
|
except Exception as e:
|
|
print(f" Failed {case['name']}: {e}")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
test_instruct()
|
|
|